How does phonon generation influence AlGaN/GaN HFETs?- Transient and steady state studies

AlGaN/GaN HFETs are high power device where power dissipation reaching more than 20 W/mm is not uncommon. What is the role of self-heating (equilibrium and non-equilibrium) and how does it influence device characterization and device performance? Traditionally the self-heating effect is assumed to be removed by using very short pulse current measurement and the fitting of hot phonon life time is then calculated based on this assumption[l], [2]. In order to check this assumption, we have developed two dimensional(2D) time dependent thermal conduction equation coupled with our 2D Poisson and driftdiffusion equation solver to study the temperature evolution in time domain for the nitride HFET. The time period between Ins-200ns and steady state temperature variation have been studied. The study shows self-heating effect may not be easily removed even with very short pulse length( 3ns). Our result explains part of the reason why low velocity still observed in the velocity-electrical field(v-E) measurement with very short pulse. The relation of current, power and temperature with time evolution will be addressed in this paper. To study the time dependent self-heating effect, we need to include the two-dimensional(2D) time dependent thermal conduction equation into our theoretical model. The thermal conduction equation is coupled into our developed 2D finite element (FEM) Poisson and continuity equation[3] to solve the current and potential self-iteratively at each time step. Figure l(a) shows typical two terminals device used for v-E measurement. The channel length is 1.6 ,um. The GaN buffer layer is 3,um and SiC substrate is assumed to be 100,um in this simulation. The mobility model is calculated from the Monte Carlo program[4] with different temperature as shown in Fig. l(b). This mobility model is then used to calculate the current with different temperature in the channel. The materials parameters used for thermal conduction equations are listed in Table I. Figure 2 shows the simulated temperature distribution in the device for two different time steps. The GaN buffer layer thickness is 3,um. As shown in the figure, at t = 10 ns, the heat generated in the channel has not have a chance to propagate to the GaN/SiC interface. This suggests that for the short pulse period, the substrate does not play any role to assist the removal of heat. Figure 3 shows the relationships of power density and temperature increase with time. The calculation shows the heating time constants are around a few nano second ranges. The average channel temperature still increases around 20K to 70K even at 2ns pulse for different power density. The thermal resistance obtained from 2D simulation is around 8.6 K mm/W, which is suitable for wide channel width cases because it does not consider the heat dissipation parallel to the channel width. Figure 4 shows the current density and temperature versus time for different VDSS. When the drain voltage increases, We observed a rapid decrease of current at t < 5ns. The temperature in the channel increases 150K and current drops 25% at t = 5 ns for VDS is equal to 20V. The decrease rate of current at t > 100 ns becomes very small compared to the current drops within 5ns. This simulation results suggest that if we use the short pulse ranges from l,us to 100 ns, to measure the v-E curves, we might get wrong conclusion that self-heating effect is completely removed since the current does not change too much within this region. Figure 5 shows the simulated v-E for different pulse length. The result shows that even for 3ns pulse, the velocity still has significant drop especially in the high field region. For the pulse length around 200ns, which is the typical pulse length used for I-V curves measurement of AlGaN/GaN HFET, the saturation velocity goes to 1.25 x 107 cm/s. This value is very close to velocity extracted from extrinsic fT measurement and is probably part of the reasons for lower expected current. More of simulation results will be presented in this talk and compared with experimental work in the future. In conclusion, our simulation results suggest that self-heating effect can not be completely removed even at very short pulse range (3ns). Better device engineering is needed to reduce the self-heating effect. The extraction of hot-phonon life time should be calibrated by the actual channel temperature for different pulse length instead of ambient temperature.